NVIDIA and Cisco unveil a new pluggable optics form factor aimed at expanding AI networking capacity

NVIDIA and Cisco announced a new pluggable optics form factor designed to help data center operators scale “AI factories” without changing the physical networking footprint. The companies said the approach is meant to let AI builders add more compute while using networking equipment more efficiently in ultra-dense deployments.

In the announcement carried by Yahoo Finance, the focus is on transceivers, which are the optical hardware modules that connect servers and switches through fiber. The new form factor is positioned as a way to “scale” within existing network layouts, suggesting customers can expand capacity without redesigning rack or interconnect plans.

The report also ties the hardware to power efficiency. Networking gear can consume a meaningful share of a data center’s electricity budget, especially in dense environments. NVIDIA and Cisco said the new design reduces networking’s share of the power budget, a framing that points to operational cost pressures as well as limits on how much heat and power the physical plant can support.

While the announcement emphasizes scaling and efficiency, it does not, in the available text, spell out performance targets such as bandwidth per port, power draw per module, compatibility with specific switch families, or a target timeline for general availability. Those details are typically crucial for operators deciding whether a new optics design can be rolled out across an existing network.

For NVIDIA, the networking push aligns with its broader push deeper into data center infrastructure, where GPUs and the surrounding interconnect fabric increasingly determine system-level throughput. The value proposition of AI “factories” is not only faster compute, but also the ability to move data quickly enough to keep accelerators busy, which raises the importance of optics, switching, and rack density.

Cisco, meanwhile, has long played in the data center networking market. By partnering with NVIDIA on a specific optics form factor, Cisco is effectively addressing a bottleneck that shows up when AI clusters expand: networking hardware must scale in capacity and efficiency while fitting within the same constrained space and power envelopes.

Industry observers often describe this as a “system bottleneck” problem: even if GPUs deliver the expected performance, the training and inference workloads can be slowed if the network cannot deliver traffic quickly enough or if adding more links forces operators to expand power delivery and cooling. The companies’ emphasis on staying within the same footprint and cutting networking’s power share suggests they are targeting those constraints directly.

What remains unclear from the information provided so far is how broadly the new modules will be supported, which vendors’ systems and specific hardware platforms will be compatible, and what the roadmap looks like for deployments. The announcement also does not indicate pricing or volume commitments. Investors and customers will likely look for follow-on technical documentation and integration guidance.